Change Detection for Multi-Polarization SAR

Appendices 89 A 89 B 93 C 95 D 97 Bibliography 101 Introduction A topic of great interest in the Remote Sensing, Signal Processing, and Synthetic Aperture Radar (SAR) communities is change detection. This is the ability to identify temporal changes within a given scene starting from a pair of co-registered SAR images representing an area of interest [1–3]. Incoherent and coherent change detection are the two main approaches that have been proposed in the open literature to process the image pair. The former attempts to detect changes in the mean power level of a given scene exploiting only the intensity information from the available images (thus neglecting phase information [4–6]): dif-ferencing and rationing are well-known techniques in this context [7]. The latter jointly uses both amplitude and phase from the reference and the test data to detect possible changes in the region of interest. In [4,5], a thorough comparison between incoherent and coherent change detection strategies, including the Maximum Likelihood Estimate (MLE) of the SAR coherence parameter, is performed based on high resolution (0.3 m×0.3 m) SAR images. In [7], several techniques for change detection have been presented and compared based on their probability of error and on results obtained using repeat-pass ERS-1 SAR data. In [6, 8–10], the multi-polarization signal model for the SAR change detection problem is laid down, the detection problem is formulated as a binary hypothesis test, and a decision rule based on the Generalized Likelihood Ratio Test (GLRT) is developed. Moreover, the performance analysis [9] of the GLRT is given in the form of Receiver Operating Characteristics (ROC), namely detection Probability (P d) versus false alarm Probability (P f a), quantifying the benefits of the multi-polarization information in SAR change detection. A complementary approach to the GLRT is considered in [11] based on the use of perturbation filters and a separated treatment between polarimetry and amplitude. A detection scheme based on canonical correlations analysis is applied III IV Introduction to scalar EMISAR data in [12, 13], whereas, in [14], a mutual information based framework is developed to address coherent similarities between multichannel SAR images. Starting from the multi-polarization data model developed in [8] and [9], a new and systematic framework for change detection based on the theory of invariance in hypothesis testing problems [15, 16] is proposed. This is a viable mean to force some desired properties to a decision statistic at the design …